YouTube Transcript Scraper (timestamps, channels, search)

Get the subtitle text of YouTube videos as one flat row per video — the full transcript and the timed segments behind it. Feed it single videos, whole channels or a keyword search. No API key, no login, no video downloads.

What you get

id / videoId, url, title, channelId, channelName, durationSec, language, languageName, isAutoGenerated, availableLanguages, segments (start, duration, text — seconds), text (all segments joined), wordCount, segmentCount, source (url / channel / search), query, scrapedAt.

Example output

Real row from the example dataset (transcript text truncated here):

Field	Example value
url	https://www.youtube.com/watch?v=dQw4w9WgXcQ
title	Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster)
channelName / channelId	Rick Astley / UCuAXFkgsw1L7xaCfnd5JJOw
language / languageName	en / English
isAutoGenerated	false
availableLanguages	["en", "de-DE", "ja", "pt-BR", "es-419"]
segments	[{"start": 1.36, "duration": 1.68, "text": "[♪♪♪]"}, …]
wordCount / segmentCount / durationSec	487 / 61 / 213

Use cases

• Feed a RAG pipeline or build a training dataset from talks, lectures and podcasts, with timestamps kept for citation.
• Summarise or translate a channel's back catalogue without downloading a single video file.
• SEO, content research and accessibility checks: read what a video actually says, in bulk.

How it works

1. Video URLs / channels / searches are resolved into a list of video ids.
2. For each video the caption track list is fetched and one track is chosen by your languages order, human-written before auto-generated (preferManual), with fallbackAnyLanguage as the last resort.
3. The chosen track is downloaded as timed text and split into segments; text is those segments joined.
4. YouTube shows a "sign in to confirm you're not a bot" wall to some exits, so the actor starts without a proxy and escalates one rung at a time (tier: auto).

Input

Field	Meaning	Default
videoUrls	Watch URLs, youtu.be / /shorts/ links, or bare 11-char ids	two example videos
channels	@handle, channel URL or UC… id — latest videos are transcribed	empty
searches	Keywords; the videos of the first result page are transcribed	empty
languages	Preference order of caption languages, e.g. ["en","de"]	["en"]
preferManual	Human-written track before the auto-generated one	true
includeAutoGenerated	Allow YouTube's speech-recognition track	true
fallbackAnyLanguage	Take any track when none of your languages exists	false
maxVideosPerChannel	Latest videos taken from each channel	50
includeShorts	List the Shorts tab instead of Videos	false
tier	Proxy rung: auto, none, datacenter, residential	auto
maxItems	Stop after this many transcripts	20
mode	scrape or monitor (only new/changed since last run)	scrape
monitorKey, webhookUrl, telegramBotToken, telegramChatId	Monitor-mode state key and alert targets	empty

Pricing

Event	Price
result	$0.002 per transcript ($2 per 1,000)
monitor-check	$0.006 per monitor run
change	$0.002 per new/changed transcript

Videos without a usable caption track are not pushed and not charged. They are counted in the ADAPTER_STATS record of the run's key-value store (noCaptions, unplayable, loginWalls, listingWalls, escalations, tierUsed), next to the standard RUN_SUMMARY record — so an empty run tells you whether the videos simply had no captions or whether YouTube refused the requests.

Why this actor

• No API key, no login, no cookies — nothing to register and nothing to keep alive.
• Timed segments come with every row, not just a wall of text, so you can cite or seek back into the video.
• A video with no usable captions is never pushed and never charged, so a batch of mixed videos costs only what it produced.
• The run tells you why it was empty: ADAPTER_STATS separates "no captions" from "YouTube wall" from "video unplayable".
• Starts with no proxy and escalates only when YouTube actually walls the exit, so most runs carry no proxy cost.

Limits

• Auto-generated tracks are speech recognition: no punctuation to speak of, occasional wrong words, and word timings grouped into short lines. isAutoGenerated tells you which kind you got.
• No machine translation: you get the tracks the video actually publishes, never a translated one.
• Videos with captions disabled, private, removed or age-restricted videos return nothing (and cost nothing).
• YouTube shows a "sign in to confirm you're not a bot" wall to some exit IPs. The actor starts without a proxy and escalates to datacenter and then residential proxy when it sees the wall; residential traffic is billed by Apify on top of the per-result price.
• Member-only and unlisted videos are not supported.

FAQ

Does it need a YouTube API key or a login? No. There is no key field and no cookie field. Captions are read from the public player payload.

What happens to videos that have no subtitles? They are skipped: no row, no charge. The count lands in noCaptions inside the ADAPTER_STATS record of the run's key-value store.

Can I get a transcript in a language the video does not publish? No. The actor never machine-translates. Ask for your languages in languages, and set fallbackAnyLanguage to true if you would rather have the video's own language than nothing.

Changelog

• 0.1: initial release — transcripts from video URLs, channels and search; language preference, manual/auto track choice, timed segments.

---

If this actor saved you time, a short review on its Store page genuinely helps other people find it. Found a bug or need a field that is missing? Open a ticket on the Issues tab.